Speaker recognition method and apparatus

1. A speaker recognition method comprising: receiving a first voice signal of a speaker; generating a second voice signal by enhancing the first voice signal through speech enhancement; generating a multi-channel voice signal by associating the first voice signal with the second voice signal; and recognizing the speaker based on the multi-channel voice signal, wherein generating the multi-channel voice signal comprises: extracting, from the first voice signal, a first feature vector including first voice information of the speaker; extracting, from the second voice signal, a second feature vector including second voice information of the speaker; and generating the multi-channel voice signal by associating the first feature vector with the second feature vector.

2. The speaker recognition method of claim 1 , wherein generating the second voice signal comprises one or both of: enhancing the first voice signal by removing a noise signal estimated from the first voice signal; and enhancing the first voice signal by increasing energy corresponding to a voice of the speaker detected from the first voice signal.

3. The speaker recognition method of claim 2 , wherein enhancing the first voice signal by removing the noise signal comprises at least one of: removing the noise signal by performing stationary noise suppression through minimum pooling on the first voice signal; removing the noise signal through channel normalization on the first voice signal; and removing the noise signal through sound source separation on the first voice signal.

4. The speaker recognition method of claim 1 , wherein extracting the first feature vector comprises: extracting a variable-length first feature vector from the first voice signal; and extracting a fixed-length first feature vector including the first voice information of the speaker from the variable-length first feature vector.

5. The speaker recognition method of claim 4 , wherein extracting the fixed-length first feature vector comprises: extracting the fixed-length first feature vector corresponding to a neural network that is trained to recognize the speaker from the variable-length first feature vector.

6. The speaker recognition method of claim 1 , wherein extracting the second feature vector comprises: extracting a variable-length second feature vector from the second voice signal; and extracting a fixed-length second feature vector including the second voice information of the speaker from the variable-length second feature vector.

7. The speaker recognition method of claim 6 , wherein extracting the variable-length second feature vector comprises: extracting the variable-length second feature vector using one or both of a feature extraction method based on a spectrum of the second voice signal, and a feature extraction method based on a neural network configured to extract a valid voice frequency interval based on a sync function of the second voice signal.

8. The speaker recognition method of claim 6 , wherein extracting the fixed-length second feature vector comprises: extracting the fixed-length second feature vector corresponding to a neural network that is trained to recognize the speaker from the variable-length second feature vector.

9. The speaker recognition method of claim 1 , wherein the multi-channel voice signal corresponds to a same utterance point as the first voice signal and the second voice signal, and includes a same dimension as the first voice signal and the second voice signal.

10. The speaker recognition method of claim 1 , further comprising: determining whether to use multiple channels, wherein generating the second voice signal comprises: generating the second voice signal based on a first determination of whether to use the multiple channels.

11. The speaker recognition method of claim 10 , wherein determining whether to use the multiple channels comprises: determining whether to use the multiple channels based on at least one of an operational load and a response speed according to a requirement of a speaker recognition apparatus, a magnitude of stationary noise included in the first voice signal, and a voice volume of the speaker corresponding to the first voice signal.

12. The speaker recognition method of claim 10 , wherein generating the multi-channel voice signal comprises: generating the multi-channel voice signal based on a second determination of whether to use the multiple channels.

13. The speaker recognition method of claim 1 , wherein generating the multi-channel voice signal comprises: determining a number of multiple channels; and generating the multi-channel voice signal by associating the first voice signal with the second voice signal based on the number of the multiple channels.

14. The speaker recognition method of claim 13 , wherein determining the number of the multiple channels comprises: determining the number of the multiple channels based on one or both of a feature of the first voice signal and noise at a location at which the first voice signal is uttered.

15. The speaker recognition method of claim 14 , wherein the feature of the first voice signal comprises: at least one of a magnitude of stationary noise included in the first voice signal, a voice volume of the speaker corresponding to the first voice signal, and a magnitude of additive noise corresponding to the first voice signal.

16. The speaker recognition method of claim 1 , wherein receiving the first voice signal comprises: collecting the first voice signal through a voice signal collector including a microphone.

17. A non-transitory computer-readable storage medium storing instructions that, when executed by a processor, cause the processor to perform the speaker recognition method of claim 1 .

18. A speaker recognition method comprising: receiving a first voice signal of a speaker; generating a second voice signal by enhancing the first voice signal through speech enhancement; generating a multi-channel voice signal by associating the first voice signal with the second voice signal; and recognizing the speaker based on the multi-channel voice signal, wherein recognizing the speaker comprises: outputting a feature vector corresponding to the speaker by applying the multi-channel voice signal to a neural network; calculating a similarity score based on a result of comparing the feature vector and a registered feature vector of the speaker; and recognizing the speaker based on the similarity score.

19. A speaker recognition apparatus comprising: a microphone configured to collect a first voice signal of a speaker; and a processor configured to generate a second voice signal by enhancing the first voice signal through speech enhancement, generate a multi-channel voice signal by associating the first voice signal with the second voice signal, and recognize the speaker based on the multi-channel voice signal, wherein the processor is configured to generate the multi-channel voice signal by extracting, from the first voice signal, a first feature vector including first voice information of the speaker, extracting, from the second voice signal, a second feature vector including second voice information of the speaker, and generating the multi-channel voice signal by associating the first feature vector with the second feature vector.